Weight Rotation Research Articles

Weight Rotation as a Regularization Strategy in Convolutional Neural Networks.

Convolutional Neural Networks (CNN) have become the gold standard in many visual recognition tasks including medical applications. Due to their high variance, however, these models are prone to over-fit the data they are trained on. To mitigate this problem, one of the most common strategies, is to perform data augmentation. Rotation, scaling and translation are common operations. In this work we propose an alternative method to rotation-based data augmentation where the rotation transformation is performed inside the CNN architecture. In each training batch the weights of all convolutional layers are rotated by the same random angle. We validate our proposed method empirically showing its usefulness under different scenarios.

Multi-Dimensional Comprehensive Link Stability Model in Ad Hoc Network

Routing Optimization is always a key issue in MANET (Mobile Ad-Hoc Network). When referring to multi-media communication, the routing stability, which is up to link stability, is in urgent need. On the analysis of kinds of factors that affect link stability greatly, a multi-dimensional comprehensive evaluation model (MVCLS) was presented based on factor classification. In the model, the weight levels were classified, a mechanism of weight rotation was available, and a normalized link stability evaluation system with multi-attribute and multi-weight was formed. In addition, aiming at reliability and robust of the model, time series analysis and Kalman filter were combined to smooth and predict the value of key factors. At last, the link stability model is used to optimize the OLSR routing protocol, which takes the link stability as the important stuff for path selection. Simulation experiments of modified OLSR were done with support of Qualnet, the results of which show that the network performance is improved and also prove the effectiveness of the link stability model.

Techniques of Transmitting Beamforming to Control the Generated Weights

In this paper, we consider the limited feedback Transmitting Beamforming for MISO systems. In conventional techniques, all vectors of a large codebook (CB), used for the feedback of the quantized channel state information (CSI), are broadcasted to all users, in a guard period which is followed by data burst periods. Instead of transmitting a high number of codevectors, we thought to divide the CB into several sub-codebooks (SC) and the broadcast would be based on the switch between them. Accordingly, a good performance can be provided while minimizing the required feedback channel capacity applying some proposed techniques such as “the switched Sub Codebook (SSC)” and “the Fairness SSC (FSSC)”. To minimize the quantization error, we propose two other techniques. The first is based on making Transmit SSC vectors controlled by a rotation weight (RW) to obtain almost a zero correlation between the SSC vectors used for the selected spatial channels. The second is based on introducing “the Schmidt algorithm” to construct an orthonormal weights using the generated weights. These two proposed techniques increase the probability of the selection of the worst case user on his best codevector to make zero the angle between his couple codevector and channel response. To analyze and validate the performance of these proposed techniques, simulation results are presented.